The brief is the specification, and it is written before the build. Records are reconciled across sources before anything is counted. Columns without a written meaning are left out rather than shipped undocumented.
Competitor screening
Removing organisations you would rather not approach, from a list you define.
Change log
A note of what each processing stage altered, kept so the build can be retraced.
Segment design
Deciding the cuts with you, against what the data can actually support.
A conversation about criteria is usually shorter than one about results.